Improvement And Application For A Class Of Compound Fuzzy Neural Network

This paper researched on a class of compound fuzzy neural network and made a variety of improvements. Firstly, improved BP algorithms are applied to recurrent compound fuzzy neural network (RCFNN), such as adding self adapting BP algorithm and momentum BP algorithm. The simulation results showed that network approximation velocity of adding self adapting BP algorithm and momentum BP algorithm are faster than traditional BP algorithm. Secondly, through research on dynamic performance of RCFNN, it is found than adding multi-layer recurrent parts improve the dynamic approximation of RCFNN. Thirdly, in industrial process, there are many prior knowledge and the rule network of compound fuzzy neural network can amalgamate these prior knowledge. Therefore, this paper researched on amalgamation of prior knowledge by rule network of RCFNN. Fourthly, this paper researched on data noise restraint ability of multi-layer recurrent compound FNN. The combination of FNN and predictive control is an effective method to improve system robustness, overcome system uncertainty and solve the control problem of uncertain system. In this paper, according to system dynamic characteristics, fuzzy predictive control is designed with multi-layer recurrent compound FNN and the simulation research show it is useful.For complex object with regions character, a method that used respective adaptive network to realize approximation in different regions is presented, which combined a number of small networks to approximate complex process object. Each network is a small self-adaptive FNN and its structure changed with different object regions.